- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hi,
I am running sample performance code from following Intel article
http://www.intel.com/cd/ids/developer/asmo-na/eng/192313.htm
This code calculates the average time taken by function ippsFFTFwd_RToPack_32f by calling it 100000 times.
Now, I tried forcing a single processor version using either w7_ a6_ and px_ version of function on same machine. On my suprise, px_version always runs 50% faster than w7_ or a6_ version. Can someone explain why so?
I am running it on my laptop
System Specification:
Centrino 1.6 GHz,
512 MB RAM,
Windows XP SP2,
MS Visual Studio .net
Clock throtteling OFF
-Visu
I am running sample performance code from following Intel article
http://www.intel.com/cd/ids/developer/asmo-na/eng/192313.htm
This code calculates the average time taken by function ippsFFTFwd_RToPack_32f by calling it 100000 times.
Now, I tried forcing a single processor version using either w7_ a6_ and px_ version of function on same machine. On my suprise, px_version always runs 50% faster than w7_ or a6_ version. Can someone explain why so?
I am running it on my laptop
System Specification:
Centrino 1.6 GHz,
512 MB RAM,
Windows XP SP2,
MS Visual Studio .net
Clock throtteling OFF
-Visu
Link Copied
4 Replies
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Hi,
something strange, could you please specify which version of IPP do you use?
Note, you can verify performance of IPP functions with performance tool which is delivered with IPP installation, take a look on IPP oolsperfsys folder where you can find performance data on different architectures as well as precompiled performance tests executables.
Regards,
Vladimir
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
it is IPP 4.1.3.
Actually, I already checked perfsys data before posting my message. As per perfsys data, w7_/a6_ version should run faster but that appears to be not true on Centrino.
On the other hand, I tried running it on system with PIII and there a6_ version running faster than px_ version.
I tried multiple times on both setup with consistent results. On PIII, a6_ version is faster while on Centrino px_ version is faster than both w7_ and a6_ version.
Indeed strange.
-Visu
Actually, I already checked perfsys data before posting my message. As per perfsys data, w7_/a6_ version should run faster but that appears to be not true on Centrino.
On the other hand, I tried running it on system with PIII and there a6_ version running faster than px_ version.
I tried multiple times on both setup with consistent results. On PIII, a6_ version is faster while on Centrino px_ version is faster than both w7_ and a6_ version.
Indeed strange.
-Visu
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Visu,
thanks for informing about that issue. We will investigate possible reasons, if you can attach here simple test case (just to let us to use the same sequence of IPP calls) it would help.
Regards,
Vladimir
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Thanks Vladimir,
I am using exact code from following article:
http://www.intel.com/cd/ids/developer/asmo-na/eng/192313.htm?prn=Y
-Visu
I am using exact code from following article:
http://www.intel.com/cd/ids/developer/asmo-na/eng/192313.htm?prn=Y
-Visu
Message Edited by visual86@yahoo.com on 08-26-2005 07:00 AM
Reply
Topic Options
- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page